I am using PCRE (same with .Net) and I would have expected the following pattern to fail when i run it against the text below:

(?<B><)?[^%>]+(?(B)>)%

that is: look for a < , look for at least 1 char which is not % or > then, if a < was found look for a >, then look for a %

in other words find expressions like ....% and <.....>%, if a < is there then there MUST be  a >

but when i run it on " <dsa z% d" it does not fail and finds the 1st 7 chars to match, what am i missing?

thanks

So, let's simplify it like this

(<)?[^%>]++(?(1)>)%

It's the same but I put posessive quantifier ++, instead of + 

Apply this to

<didi%

The regexp engine do this:

- first it tries for "<", will match

- then will match "didi"

- then, cause it find the first "<" will try to match ">", but it fails, so the engine backtrack again to the first "<", now because it's optional, the "<" will be matched by [^%>]+ part, so the conditional part (?(1)>) will not match (it will skip that part because first group, B in your case, isn't match after backtracking)

- then will match %

Of course, silly me. I should have used  "^ *(?<B><)?[^%><]+(?(B)>)%"

Thanks

Use posessive quantifier [^%><]++ OR atomic grouping (?>[^%><]+) for that part, cause is more efficient that way in the case of non matching